A NNALES DE L ’I. H. P., SECTION B
C ARLO B OLDRIGHINI
R OBERT A. M INLOS
A LESSANDRO P ELLEGRINOTTI Interacting random walk in a dynamical random environment. II. Environment from the point of view of the particle
Annales de l’I. H. P., section B, tome 30, n
o4 (1994), p. 559-605
<http://www.numdam.org/item?id=AIHPB_1994__30_4_559_0>
© Gauthier-Villars, 1994, tous droits réservés.
L’accès aux archives de la revue « Annales de l’I. H. P., section B » (http://www.elsevier.com/locate/anihpb) implique l’accord avec les condi- tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti- lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
Interacting random walk
in
adynamical random environment
II. Environment from the point of view of the particle
Carlo BOLDRIGHINI
(*)
Robert A. MINLOS
(*)
Alessandro PELLEGRINOTTI
(*)
Dipartimento di Matematica e Fisica, Universita di Camerino, Via Madonna delle Carceri, 62032 Camerino, Italy.
Moscow State University, Moscow, Russia.
Dipartimento di Matematica, Universita di Roma La Sapienza,
P. A. Moro 2, 00187 Roma, Italy.
Vol. 30, n° 4, 1994, p. 559-605. Probabilités et Statistiques
ABSTRACT. - We
consider,
as inI,
a random walkXt
E EZ+
and a
dynamical
randomfield ~ (x),
x E Z" in mutual interaction with each other. The model is aperturbation
of ununperturbed
model in which walk and field evolveindependentently.
Here we consider the environment process in a frame of reference that moves with thewalk, i.e.,
the "field from thepoint
of view of theparticle" ~t (.) == çt (Xt +.).
We prove that its distributiontends,
as t -~ oo, to alimiting
distribution tc, which isabsolutely
continuous with
respect
to theunperturbed equilibrium
distribution. We also provethat, for v > 3,
the time correlations of the fielddecay
ase-cxt
const
20142014.
.Key words: Random walk in random environment, mutual influence, environment from the
point of view of the particle.
(*) Partially supported by C.N.R. (G.N.F.M.) and M.U.R.S.T. research funds.
A.M.S. Classification : 60 J 15, 60 J 10.
Annales de l’lnstitut Henri Poincaré - Probabilités et Statistiques - 0246-0203
Vol. 30/94/04/$ 4.00/@ Gauthier-Villars
560 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
On considere un chemin aleatoire
Xt
E EZ+
etun milieu aleatoire
dynamique çt (x), x
E Z" en interaction mutuelle.Le modele est une
perturbation
d’un modeleimperturbe,
danslequel
le chemin aleatoire et le milieu evoluent d’une
façon independante.
Onetudie ici le
proces
du milieu dans unrepere qui
sedeplace
avec lechemin
aleatoire,
c’ est-a-dire le « milieu dupoint
de vue de laparticule »
r~t
( . ) (Xt +.).
On montre que la distributionde T/t tend, pour t
-~ o0a une distribution limite
qui
est absolument continue parrapport
a la distributiond’ équilibre
du milieuimperturbe.
On montre aussi que lepremier
terme dudeveloppement asymptotique
des correlationstemporelles
du
champ ~t
est donne par constt2 " e-at
.1. INTRODUCTION
The
present
paper is second in a series of two papers. As in thepreceding
paper
[1] (hereafter
referred to as PartI),
westudy
the time evolution of arandom walk
Xt
on the v-dimensional lattice Z" and a field(environment) çt (x) :
x Esubject
to a mutual interaction of local character.Time is
discrete, t
EZ+,
and the field E Z" takes values ina finite set S.
We
briefly
recall the main features of themodel,
and refer the reader to Part I for more details. Theassumptions
of thepresent
paper which differ from those of Part I are listed in Section 2.As a
starting point
we consider an"unperturbed" model,
in which the random walk and the environment evolveindependently.
The random walk ishomogeneous
with transitionprobabilities ~ Po (y) :
y E7~v ~,
and theevolution of the environment at each site is an
ergodic
Markovchain,
with afinite space state
S,
and stochasticoperator Qo = ~ qo (s, s’) : s,
s’ ES ~,
the same for all sites. 7ro will denote the
unique stationary
measure of thechain. The evolution at different sites is
independent.
For the
interacting
model the random walk transitionprobabilities
arewritten as
Annales de l ’lnstitut Henri Poincaré - Probabilités et Statistiques
and the transition
probabilities
for the environment at the site x E Z" arec ( . , . )
andq ( . , . ) satisfy
somecompatibility conditions,
and e is a smallparameter.
Under some
general assumptions
we deduced in the paper[2]
the localcentral limit theorem for the
displacements
of theparticle.
In Part Iwe studied the
asymptotic (in time) decay
of the correlations of the environment in a fixed frame of reference. Thepresent
paper is devoted to theinvestigation
of the environment process in a frame of reference whichmoves with the
particle, i.e.,
of theprocess {~t : t
EZ+}
defined asqt
(x) _ ~t (Xt
+x) (sometimes
called "field from thepoint
of view of theparticle").
~t evolves as an infinite-dimensional Markov chain.We prove for any dimension v
that,
as t ~ oo, the distributionof ~t
tends to a
limiting
invariant distribution p, which isabsolutely
continuouswith
respect
to the invariant distributionIIo
= of the environment in theunperturbed
case(e
=0).
We alsostudy
the timeasymptotics
of the(time)
correlations of the field forv >
3. We prove that theleading
term is of the
type
conste-at t- 2 .
The constant factordepends
on theinitial
conditions,
whereas a E(0, oo ) depends only
on theparameters
of the model. This result should becompared
with thelong
time tail of the correlations of the field in a fixed frame ofreference,
which was studied inPart I. The different behavior is
explained by
the fact that in a fixed frame of reference the environmentprocess 03BEt by
itself is not Markov.The methods used in the
proofs
are, as in theprevious
papers[1] ]
and[2],
based on thespectral analysis
of the stochasticoperator (or
transfermatrix) T
of the Markovchain {~t : t
EZ+}, acting
on the space=
L2 (H, IIo).
Here H = is the state space of the field. The method of theproof
isbased,
as in PartI,
on theanalysis
of theleading spectral subspaces.
The paper is
organized
as follows. Section 2 is devoted to the definition of the model and to the statement of the results. In Section 3 we prove the main technical theorem(Theorem 3.1),
whichgives
thedecomposition
of J-lin invariant
(with respect
toT) subspaces.
Section 4 contain theproofs
ofthe
theorems,
whichrely
on the results of Section3,
and Section 5 is devoted to someconcluding
remarks. In theAppendix
we prove a technical lemma which is needed in theproof
of Theorem 3.1.Vol. 30, n° 4-1994.
562 C. BOLDRIGHIM, R. A. MINLOS AND A. PELLEGRINOTTI
2. DEFINITIONS AND FORMULATION OF THE RESULTS The model is described in detail in Section 2 of Part
I,
to which we refer. We state hereonly
theassumption
which differ from the ones ofPart I.
Throughout
the paper we will write(I n.m)
to denote formula(n.m)
of Part I.
The which
plays
animportant
role in whatfollows,
is definedby (I 3.1 a).
Throughout
the paper we assume conditionsI,
II and III of Part I onthe random walk transition
probabilities,
and conditions IV and V on the transitionprobabilities
of the random field. Condition VI on thespectrum
ofQo
isreplaced by
thefollowing
one.VI*.
If I S ]
> 2 thespectrum
ofQo
is such thatCondition VI* is the old condition VI
plus
theassumption
that theeigenvalue
isnondegenerate (hence real,
sincecomplex eigenvalues
occur
only
inconjugate pairs).
Further,
we assume Condition VII of Part I for the function c, butreplace
Condition VIII
by
thefollowing (stronger)
condition.VIII*. The Fourier coefficients
of the inverse of the function
po (~) [eq. (I
2.3d)] satisfy
theinequality
Inequality (2.1 b) implies
the existence of aspectral
gapWe now define the random field which will be studied in this paper, the
"environment from the
point
of view of theparticle".
This is the random field qt E EZ+ given by
the relationAnnales de l’Institut Henri Poincaré - Probabilités et Statistiques
For any bounded functional F on S2 we have for the conditional average
[with respect
to the distribution(I 2.1)]
where
Py (. ~
is aproduct
measureqI
( s, s’ ) ,
s,s’
E Sbeing
the matrix elements of the matrixQi
definedin eq.
(I 2.5b).
The conditional average(2.2b)
does notdepend
onXt-i,
and we can consider it as an average over the Markov
field {
1]t(y) },
withtransition
probabilities
The stochastic
operator
or transfer matrix T of this field isdefined,
asusual, by
its action on the bounded functionals F on H :We denote the average of a random variable
f
withrespect
to anymeasure v
by ( f )",
and the correlationsby ( f, g ~" _ ( f g ~" - ( f )" ~ 9 ) ~.
We fix an initial distribution II of the field ~, induced
by
some initialdistribution II of the
field ç
for a fixed initialposition
xo =Xo
of therandom walk. II is
simply
the shift offI by
xo.Pn
denotes the distributionon the space of
trajectories
of the Markov field{ r~t :
t E7L+ } generated by
the initial distribution II. From the definition
(2.3)
wefind,
for F as beforeWe shall also consider condition IX of Part I
(symmetry
of the randomwalk).
For thefield 17
itimplies
thefollowing
statement, which may be considered as a new condition.Vol. 30, n° 4-1994.
564 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
IX*.
Here,
as in PartI,
V is the space reflection:(V ~7) (x) _ ~ (-:c).
We need the
following
furtherassumption
on the correlations of the initial measure II.X*. The
spatial
correlations withrespect
to the distribution II of the vectors{03A80393:
r E of the basis in H[see (I 3.1a)] satisfy
thefollowing
clusterinequality
Here denotes the multi-index with supp r
= ~ x ~
~y(x)
=j,
andM, q
are two constants such that M >1,
andq
E(0, 1).
As in Part
I, by dA,
A c we denote the minimallength
of theconnected
graphs which join
allpoints
of the set A.We now formulate the main results of our paper. We understand
throughout
that condition I-IX are the ones of Part I. Conditions introduced in thepresent
paper are denotedby
a star.THEOREM 2.1. - Let t E
l~+
denote thefamily o,f’measures generated by
the Markov process r~t with =II,
and assume conditionsI-V, VI*,
VII and VIII*.
Then,
as t ~ ~, the measures tendweakly
to aninvariant measure p, which is
absolutely
continuous withrespect
to theindependent
measureno.
Moreover there are
positive
constants cl andq
E(0, 1)
such thatRemark 2.1. -
Inequality (2.4) implies
that Condition X* above holds for the invariant measure J1 as well.In Theorem 2.1 Condition VI* could be
replaced by
the weaker ConditionVI,
at the cost of alonger proof.
We shall indicate below how it can be done.The other result concerns the behavior of the time correlations of the
field,
Annales de l ’lnstitut Henri Poincaré - Probabilités et Statistiques
for which we write down the
precise asymptotics.
Consider the correlationwhere x1, x2 are two fixed
points.
THEOREM 2.2. - Let
v > 3,
and add conditions IX* and X * to thehypotheses of
Theorem 2.1.Then,
as t --~ oo,where a > 0 is a constant
depending only
on theparameters of
the modeland the constant C
depends
in addition onf 1, f 2 ,
x 1, x2, and II.By
Remark 2.1 the invariant measure ~c satisfies the conditions of Theorem2.2,
so that theequilibrium
time correlations for the measure pare also of the
type (2.6).
3. EXISTENCE AND PROPERTIES OF THE INVARIANT SUBSPACES
3.1. More notation and
preliminary
resultsIn what follows we denote
by
const any absolute constant,independent
of c.
The action
of T
on the basisfunctions {03A80393 : 0393 ~ 203C0} is given by
Here
Aj (x)
r is the multi-index obtainedby replacing
thevalue, (x)
of rat x
with j, leaving
all the othervalues, (~), ~ 7~ ~ unchanged. T°
is theunperturbed operator
Vol. 30, nO 4-1994.
566 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
and the coefficients
L ( ... )
aregiven by
Here we have used the
equality Aj ( - y )
F+ y
=Aj (0) (F
+y ) ,
and thecoefficients cj
(y), qj’j ,
andbjm,
m’ are the coefficients of theexpansions
inthe
basis (
of the functions c(u, s), 03A3 q (s, s’)
ej(s’)
and ek(s) em (s) , respectively [see (I 2.6f)].
Note that co(u)
= 0and q0k
=0,
whichimplies L(0, j; u)
= Cj{u) .
We shall write T =
T°
is translation invariant whereas T is not.LEMMA 3.1. - T and 0 are bounded linear
operators
on ?-~.Proof. - By
eqs.(3.1a, b, c)
we havewhere we use the notation
Recalling
theexpression (I 3.1b)
of the scalarproduct
inH,
we haveSetting,
for fixed y, F* =T"
+ y, =f r~ _y,
we haveAnnales de l’lnstitut Henri Poincaré - Probabilités et Statistiques
where the
operator
is definedby
theposition
The
proof
thatT (y)
is a boundedoperator
is doneexactly
as in Lemma A.Iof
[2],
andsince II by
translation invariance of the norm, and the sum is over the finiteset I y ] D,
it follows that T is a boundedoperator
in H. Theproof
that A is bounded is an easy consequence.Lemma 3.1 is
proved..
As in Part I we
consider,
for M > M* -max {max|ej (s)|, 2 },
thes
dense
subspace HM
C~-l,
withnorm ] ] ~~M
definedby (I 3.5a).
By
the definition(I 3.1a)
of the basis{ ~r }
it is easy to see that thefollowing inequalities
between norms holdwhere II
msup ~/ (~) ~ ]
denotes the supremum normof f.
For the space
HM
we state theanalogue
of Lemma 3.1.LEMMA 3.2. -
(i)
Theoperators
T and 0394 are bounded onHM . (ii) HM
is invariant under T.
Proof. -
It is the same as theproof
of theanalogous
Lemma 3.4 in[2] ..
We denote
by
the set of theequivalence
classes of multi-indices which differonly by
a shift.Let (
E andF E (be
arepresentative
of the class
(.
SinceT°
is translationinvariant,
thesubspace consisting
of the vectors
is invariant under It is easy to see that
Hence the
spectrum
of theoperator x ~
coincides with the range of the function Note that does notdepend
on the choice ofthe
representative r
E(.
We denoteby x°
the space~C~ when (
is theclass of
equivalence
of the multi-indices F which contains(defined
inVol. 30, n° 4-1994.
568 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
Condition X*
of § 2).
Condition VIII*implies
that thespectrum
ofT°
in the space~C°
isseparated
from thespectrum
of2~°
in~-C B ?-~o .
3.2. Construction of the invariant
subspaces
The main result of this section is the
following
theorem.THEOREM 3.1. - Under the
assumptions I-V, VI*, VII,
andVIII*,
andfor ~
smallenough
thefollowing
assertions hold.(i)
One canfind
twosubspaces of ~-C, ~1
and~nC2,
invariant with respectto the
operator T,
such thatbeing
the spaceof
the constants.(ii)
In the space one canfind
abasis ~ hz ,
z E on which theoperator T1 ~ T
actsaccording
to theformula
Moreover
p (y)
is real andgiven by
the relationand, for
some q E(0, 1),
thefollowing inequalities
hold(iii)
The normsof
the restrictionsof
T to thesubspaces
and(endowed
with thenorm ~ ~ . ~ satisfy respectively
the estimatesiv)
Under the additional condition IX*p (y)
is an evenfunction of
y E ~".
Remark 3.1. - The
decomposition (I 3.2)
of thespace ~
in a directintegral
of theeigenspaces
of the group oftranslations {Uv : v
does not reduce the
operator T,
since for c > 0 T does not commute withAnnales de l’lnstitut Henri Poincaré - Probabilités et Statistiques
the space shifts. This is
why
an additional sum appears in(3.4a),
and wehad to assume
inequality (2.1c)
in Condition VIII*.Remark 3.2. - If we have condition
VI,
instead ofVI*, i.e.,
theeigenvalue ~ 1
has amultiplicity s
>1,
Theorem 3.1 holds with someobvious
changes. Namely
the basis in the space will be labeledby
two E
ll v , i = 1, ... , s ~ .
Moreover theoperator T |H(1)M ~ T1
in this basis will act as followsM
where
and the matrix
elements {03B4ij (u)}, and {Sij (z, y)} satisfy
estimatesanalogous
to(3.4c).
Proof of
Theorem 3.1. - Theproof
makes use of the same ideas as theanalogous proof
in section 3 of[2].
It is based on severallemmas,
whichare stated and
proved
in the rest of thepresent
section.We shall first find the invariant
subspace
and then obtainHi
asthe closure of in H.
For c = 0 we have
clearly
~C = ~-lo ~-- ~C° + ~‘~C2 ,
~i, M + ~2 ~
M,~° l,,l = ?-~C° n ~-~CM 2 = 1,
2.Here
Ho
is the space of the constants,?-~°
was defined above andHg
isthe
(closed) subspace spanned by
thefunctions {03A80393:
1 r Eg2}, g2 being
the set of the multi-indices r with
either supp h ~ ]
> 1or supp r ~ [
= 1but 03B3 (x) ~
1 for x E supp r.It is convenient for the moment to set
so that
HM
=H01,
M + M, and T(as
anoperator
on is writtenas an
operator
matrix’
Vol. 30, nO 4-1994.
570 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
where the
operators Ti~
act as follows:Tll :
~ -~ ~,,
~21 :
~-~ ~ l°~M, T22 :
Thespace , is
spanned by
the . : x Eand 03A80.
i
We first
study
the inverse of theoperator T~l
in the space l,,l.is defined in Condition X* of Section
2.)
We introduce forbrevity
thenotation ~pz
= ~ riz ~ ,
and we will allow z to take the value0 :
cpo =Wj.
For the indices of the matrix elements of
operators
on~l°
we shall also write z instead ofLEMMA 3.3. - The
operator T11
= T~~a
is invertible in and1, M ,
its inverse is
given by
,
where the
operator
D has thefollowing properties:
for
some 0 E(0, 1).
Proof. -
Theoperator 7n
can be written asand,
as it follows from formulas(3.1 a, b, c) 7n
can be written as a matrix(corresponding
to thedecomposition
=Ho
+~C°, ~ )
where
Its inverse is
represented by
the matrixwith
Annales de l’lnstitut Henri Poincaré - Probabilités et Statistiques
and R is
given by
a serieswhere in the
running
term of the sum theproduct R
isrepeated k
times. From
eq. (3.8d)
weget
constBf x-~ ~
I for some0 E
(0, 1). Inserting
theexpression
for7Z given by (3.8b)
oneeasily
finds the estimate
One finds also that
This proves Lemma 3.3..
We look for an invariant
subspace
of the formwhere S : --~
?nC°, ~
is an unknownoperator.
The invariance condition for leads tothe following equation
for S[which
isanalogous to eq. (I 3.6c)] :
JC is considered as
acting
on the spaceA (?-L°, ~ , ?nC°, ~ )
of the maps fromto
?nC°,
~, endowed with theoperator
norm.Equation (3.9b)
has aunique solution,
as statedby
thefollowing
lemma.LEMMA 3.4. -
If ~
is smallenough
one canfind
a number K0 > 0 such that the map lC is a contraction in thesphere of
radius ~o centered at theorigin of
the spaceA (~C°
M,~C° ~ ) .
Proof -
We denote withSz,
r the matrix elements of theoperator
S.By equations b, c)
we see that(~21 )o, r
=0,
and for z E 7~vFor
(T12 ) r, z
wehave,
ifsupp r = ~ ~c ~,
Vol. 30, n° 4-1994.
572 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
and if supp r
= ~ u ~
U{ v },
and z E Z"= 0 in all other cases.
For the norms of the
operators appearing
inequation (3.9b)
we havewhere the constants
Ci, j, i, j
=1,
2depend
on theparameters
of the model and on M.(3.12a)
comes from Lemma3.3, by observing
thatInequalities (3.12b, c)
areeasily derived, using
theexplicit expressions (3.10), (3.1 la, b)
and(3.12d)
followsby observing
that for ~ =0, ?22
with ( T° ~ ~ _
~c*, and that~-1 (~22 -
is a boundedoperator (Lemma 3.2).
By inequalities (3.12a, b,
c,d)
and Condition VIII*[inequality (2.1b)],
we see that the second term on the
right-hand
side of eq.(3.9b)
isbounded,
for small c,by ~3 ( ( S ( ~,
where/3
E(0, 1).
Since the other two terms areof order c
and ~ ~S~ 112 respectively,
K is a contraction in anysufficiently
small
sphere
of Lemma 3.4 isproved..
Hence there is a
unique S,
solution of eq.(3.9b)
and the space definedby
eq.(3.9a)
is invariant withrespect
to T. Since M >1,
it is not difficult to see thatAnnales de l ’lnstitut Henri Poincaré - Probabilités et Statistiques
The Banach space of the elements of
A (~‘~C°,
~,~C2, ~)
with the norm( M
is denotedby AM.
The solution Sof equation (3.9b)
isactually
in
AM,
as it follows from thefollowing
lemma.LEMMA 3.5. - The
right-hand
sideof equation (3.9b) defines
a mapAM ~ AM,
andfor ~
smallenough
one canfind
apositive
number Rso small that
J’CM
is a contraction in thesphere of
radius ~ centered at theorigin of AM.
Proof. -
Wehave, by inequalities (3.12a, b,
c,d)
For the second term, which is the
leading
one, we havewhere
/3
E(0, 1 ) .
The conclusion follows as for thepreceding
lemma..Hence there is a
unique
solution of eq.(3.9b),
which coincideswith the
previous
one. , We introduce a basis inby setting
The action of T on the new basis is
given by
thefollowing
formulasVol. 30, nO 4-1994.
574 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
We need to prove a
property
of fastdecay
as z,z’
-~ o0 of the matrix elementsAz,
z~ . To thisaim,
we use the factthat,
as it wasproved
in[3],
S can be written as a series
where the summation goes over all sequences of
pairs
ai =(si, qj),
Si, qi E
Z+,
and x~l, ", , ar are real numbers
(see [3]).
We first prove the convergence of the series.LEMMA 3.6. - For ~ small
enough
the series(3.16a)
converges in thenorm
of AM.
Proof. - By reasoning exactly
as in the deduction ofinequality (3.13a)
we find
and,
inanalogy
with the deduction ofinequality (3.13b),
we findwhere,
asbefore, by [ [ . [ we
denote theoperator
norm of722 :
~(o~
’
2, M.
Now, by
Lemma3.3, expanding
the power(Tl 1 ) -S
=( (~° ) - ~ -f- ~ D) s,
we find
where
and
D(q)
is written as a sum ofproducts.
Thefunction
which isthe Fourier transform of
(T011)-1,
isanalytic
in somecomplex neighborhood Wd
={ A : ~ d,
z =1,... ,~ }
of the torusT",
if d issufficientely small,
and satisfies inWd
theinequality
where r
( ~c) , u
E Z" are the Fourier coefficients(2. la),
and thequantity
r~is small for small d. It is easy to see that the Fourier transforms
are
analytic
for~,
~’ EWd. Using
theinequality (1
+x)s -
1x 2
( 1
+x 2 ) s ,
we findwhere ci, c2 are constants
independent
of s.[Note that, by
formula(3.8c)
we have
( ~11 ) oo
= 1 and= 0, z
E Z" U0.] ]
Incomputing
theFourier coefficients we can now shift the
integration region
in thecomplex region Wd,
and weget
from(3.19), (3.20a, b) (maybe by redefining r~)
the relations
Relations
(3.21) imply
..u C L- .. - ..
where c is a constant
independent
of s. From(3.17), (3.12d), (3.22)
we findVol. 30, nO 4-1994.
576 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
whence
r r
where Sr
=Y~
Si andQr
=1 1
From the
analysis
in paper[3]
it follows that xai, ... , ~r = ~ if and thatwhere ws, r are the coefficients of the
expansion
of the solution w
(z, ()
of theequation
By looking
at thesingularities
in thecomplex z-plane
of the solution of thequadratic equation (3.24b)
it is not hard to see that the radius of convergence of the power series in z of w((, z)
tends to 1as (
--~ 0. Sincein our
case ~ ~ ~ _ (C ~)2,
and the number1
+ ~C22 ) a + ~ is, by
Condition
VIII*,
less than I for e and d smallenough, B conclude
thatCondition
VIII*,
less than 1 for ~ and d smallenough,
we conclude that there is apositive
number 6-0 such thatfor, s
co,(1+~ C22 ) (a + ~ |1|)
will be smaller than the radius of convergence
(in z)
of the series(3.24a),
whichimplies
absolute convergence inAM
for the series(3.16a).
Lemma 3.6 is
proved..
Remark 3.3. - Note that since and
T21 03A80
=0,
itfollows that
0,
whichimplies
~co =~o~
Remark 3.4. -
Inequality (3.23b) implies [
const ~.For what follows we need a more accurate
analysis
of the matrix elements of theoperator
S. What we need isprovided by
thefollowing
Lemma.LEMMA 3.7. - The
following
relations holdAnnales de l’lnstitut Henri Poincaré - Probabilités et Statistiques
where the
functions
R and Vsatisfy, for
some 8 E(0, 1),
the estimatesProof. -
Theproof
is ratherlengthy
andrequires
additional constructions.It is deferred to the
Appendix..
By
Lemmas 3.5 and 3.7 we haveproved
that is invariant withrespect
toT,
and that one can find in it abasis {uz : z
E Z" U0}
onwhich T acts as in
(3.15a),
and thecoefficients ~ Az, z~ : z, z’
E Z" U0 ~
are,
by (3.8a, b), (3.15b)
and(3.25a),
To
accomplish
theproof
of assertion(ii)
of Theorem 3.1 we introduce in a new basisand we show that the
sequence H
=f Hz : z
E7L" }
can be determined in such a way thatIn
fact, by
eqs.(3.15a)
and(3.27a)
we haveand we can define H to be the solution of the
system
ofequations
Let
Bq, q
E(0, 1)
be the space of the sequences H such that the normis finite. From formula
(3.15b),
Lemmas 3.3 and 3.7 it follows that for q > 6~ the z’ E defines a boundedoperator A
inBq,
with1, and, moreover {Az,0 :
z E EBq.
Vol. 30, n° 4-1994.
578 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
Hence there is a
unique
solution inBq
of thesystem (3.28a). Moreover, by
formula(3.26)
it is clearII q ~ const
whence it follows that the sameinequality
holds for the solution of eq.
(3.28a).
By (3.27b)
the spacesplits
into two invariantsubspaces:
thesubspace
of the constants1-lo
={
cho } _ ~
c~o }.and
thesubspace
which is
spanned by
thebasis {hz : z
E7L" }. Setting
nowb (y)
= R(y),
and
S (x, y)
=y)
+Vx,y, using
the estimates(3.25b)
weget
therelations
(3.4a, b, c)
of Theorem 3.1.We now prove
inequality (3.5a)
for the restriction of T toUsing
theexplicit expression (3.27a)
for the functions of thebasis { h,z },
Remark 3.4and
inequality (3.28b)
we see thatwhere the constant is
independent
of ~ and z. For any vector v =LV z h z
Ez
~CM , by susbstituting
theexplicit expression (3.27a)
of the functionshz,
we find
whence, reasoning
asabove,
onededuces,
for some ci >0,
Using (3.29), (3.30)
and theinequality
H ‘
which follows
easily
fromequation (3.26),
and Lemmas 3.3 and3.7,
wefind,
for ~ smallenough
Annales de l ’lnstitut Henri Poincaré - Probabilités et Statistiques
Inequality (3.5a)
isproved.
We are left with the construction of the space
7-lM~.
We recall that in[2]
we introduced a basis
{ 03A8*0393 :
r E in thespace H
=
0, ... , |S| - 1 }
is a basis in the spacel2 (8, given by
the normalizedeigenvectors
of theoperator
andbi-orthogonal
to the
basis {ej : j
=0, ..., |S| - 1}.
The r E isbi-orthogonal
to thebasis {03A80393
1r E
in H. We observefurthermore
that thematrix {T*0393.0393’} of
theoperator T*, adjoint of T,
in the basis r E isthe adjoint
of thematrix { 7r, r~ ~
of theoperator
T in thebasis {03A80393,
r Ei.e.,
By applying
to theoperator
T* the same considerations as above we finda
subspace
C1-lM,
invariant withrespect
toT*, which,
inanalogy
with
(3.9a),
isgiven by
Here is the space
spanned by
the vectors~po
and{ cpz
:z E
cpz = ~ ~~ z ~ ,
is defined inanalogy
to and theoperator
s* : is the space
spanned by
the vectors~ ~ r :
r E~2 ~ )
is defined inanalogy
to S. In the space we can choose a E Z" U0 ~
of the form~cz
==cp;
+ ?* It isnot hard to see that
T* acts on the basis
{ ~cz ~
as followsand,
inanalogy
with eq.s(3.26),
= 1 and = 0 for all z E Z".As above we can go over to a new basis in
and the
sequence {H*z, z
E is the solution of thesystem
Vol. 30, n° 4-1994.
580 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
which, reasoning
as for eq.(3.28a),
is in the spaceBg.
Wefind,
asabove,
i. e.,
the space* splits
into two invariant(with respect
toT* ) subspaces:
and *
being spanned by
the vectorsho,
We denote
by
and thesubspaces
of ~-C obtainedby taking
the closure of the spaces * and
7~
* in the norm of ~.They
areinvariant with
respect
to T* and for them adecomposition analogous
to(3.35)
holds.We now come to the
proof
of thereality
of the functionp (~/).
It followsfrom the
reality
of theoperator
T(i.e.,
from the fact thatT f
=T f ),
andthe
reality
of the functions( ~ )
= ei( ~ ( z ) ) : z
E andwhich,
in its turn, follows from Condition VIII*. This
implies
thereality
of theoperators 7n, 7i2, T21,
and?22.
In
fact,
consider for instance a real functionf
E?-~°,
M . As T isreal,
T
f
=?i2 f
+?22 f
isreal,
and so isT~12 f = L (T f, cpz)
~pz.Hence
7i2
and722
are realoperators. Analogous arguments
show that7n
and
Tl2
are also real. Thisimplies, by equation (3.16a),
thereality of S,
and hence the
reality
of thebasis {hz : z
E and of the matrix elementsAz,
z~ .Since,
as it follows from estimate(3.4c)
and formulas(3.26), p (y)
= limAz,
z-y, is a real function.If we now assume that the
symmetry
Condition IX*holds,
it is not hardto see that
Az, z~
=Az~,
z. Infact, by
the discussion of Condition IX in PartI, Po ( y )
and cj(y), j
=1, ... ~ S ~ -
1 aresymmetric in y
E Itfollows that the
operators 0~
definedby
formula(A.4)
of theAppendix,
are also
symmetric
in yj. Thisgives
therequired property
of the matrixAz,
z~,which,
with thehelp
of(3.26)
leads to thesymmetry
of the functionp ( y) .
Assertions(ii)
and(iv)
of Theorem 3.1 areproved.
We now pass to the rest of the
proof.
We haveAs a consequence of the
bi-orthogonality
of thebases {
r em}
and {Wr :
r Ehj
isorthogonal
to thesubspace ?-~1.
Moreoverho
is
orthogonal
to thesubspace
Infact, by
the invarianceof h*
withrespect
to T* we have( ho , (E - T) v )
= 0 for any v E?-~~ ,
where Edenotes the
identity operator. But, by inequality (3.5a) f - ~
is invertibleAnnales de l ’lnstitut Henri Poincaré - Probabilités et Statistiques
in
~‘~CM ,
whichimplies
thatho
1 v for any v E~M .
Since is densein
Hi,
the assertion follows.Next we construct the
bi-orthogonal
basisto {hz : z
E7~v ~. By
relations
(3.14), (3.26), (3.31)
and(3.32)
we haveWe introduce a new E in the space of the form
The matrix
.~ _ ~ .~’z, z~ ~
is chosen in such a way thatwhere 7* is the
adjoint
matrix of .~’.In Lemma 3.8 below we prove that the
operator
C has a small norm inBq .
Hence.~’,
which can be written asexists as an
operator
onBq,
and has also small norm.An easy check shows that the E is
bi-orthogonal
to the
basis {hz :
z E inxl.
We now introduce the space
Clearly
this space is invariant withrespect
toT,
and for any vectorf
we have a
unique decomposition
’
This last assertion can be
proved
as follows. We setIt is easy to see that
f(2)
1ho
andf(2)
1?~Ci,
so thatf(2)
EH2.
Theuniqueness
of thedecomposition (3.37a)
is evident.This proves the
decomposition (3.3a).
Toaccomplish
theproof
wehave to prove the
analogous
assertion(3.3b)
forHM.
We have to showVol. 30, n° 4-1994.
582 C. BOLDRIGHINI, R. A. MINLOS AND A. PELLEGRINOTTI
that and
f t2~
are in and~C~ , respectively,
for anyf
EHM.
Wefirst need an estimate for the matrix elements
Sz, rand Sz
r.LEMMA 3.8. - The
following
estimates holdwhere q E
(0, 1 ) , Cl
is an absolute constant, and u is the elementof
theset supp r which is closest to z. A similar estimate holds
for Sz,
r, z E7~v,
and
finally
"Proof. -
It is not hard to deduce the estimate(3.38a) by analyzing
theterms of the series
(3.16a, b),
much in the same way as it is was done in theproof
of Lemma 3.7. As forS*,
it can berepresented
in aseries,
whichis
simply
obtainedby replacing
in the series(3.16a, b)
the matrix elements of theoperator
T with the elements of theoperator T*.
The estimates thatare needed are obtained as for S. We omit the details..
To prove that the
projection
of any vectorf
EHM
onH2
andHi belongs respectively
to or~nC~ ,
andinequality (3.5b)
we introducea basis in
H2 .
Consider the functions
_ - - -
which,
as it is easy tocheck, belong
to the spaceH2.
In addition we set= uz, z e Z"
[see (3.14)],
andUj
= uj =Wj.
i
LEMMA 3.9. - The
following
assertions hold.(I)
Thefunctions ( Ur,
r eC2 )
are inH2,
M.(ii) Any
vectorf
eH M
is written as a serieswhich converges in the
and, for ~
smallenough,
we haveAnnales de l’lnstitut Henri Poincaré - Probabilités et Statistiques